31 Journal of Water and Health | 04.Suppl 2 | 2006 The

This paper reviews estimates of the incidence and prevalence of acute gastrointestinal illness (AGI) from 33 studies. These studies include prospective cohort studies, retrospective cross-sectional population-based surveys, and intervention trials from the United States and six other developed countries published since 1953. The incidence and prevalence estimates for AGI reported in these studies range from 0.1 to 3.5 episodes per person-year. However, comparisons of these rates are problematic owing to significant variation in study design, sampling methodology, and case definitions and should be made with caution. In the United States, the Centers for Disease Control and Prevention's (CDC) Foodborne Diseases Active Surveillance Network (FoodNet) estimates a rate of 0.65 episodes of AGI per person-year. This estimate includes diarrhea and/or vomiting of infectious or non-infectious origin, with a measure of severity (impairment of daily activities or diarrhea duration greater than 1 day), and has been adjusted for combined respiratory-gastrointestinal illnesses. However, it excludes episodes of diarrhea or vomiting due to any long-lasting or chronic illness or condition. Limitations in study design result in an unknown degree of uncertainty around this point estimate.

months of age experience the highest median rate of diarrhea at five episodes per person-year (Snyder & Merson 1982;Kosek et al. 2003).According to 2002-2003 World World Health Organization statistics, diarrheal diseases accounted for 17% of all deaths worldwide in children younger than 5 years of age, resulting in approximately 1.8 million deaths annually (WHO 2005).When all ages were considered, diarrheal diseases accounted for 3.2% of all deaths worldwide each year.Diarrheal diseases have the 5th highest burden of disease, expressed in DALYs, surpassed only by perinatal conditions, respiratory infections, HIV/AIDS, and depression (WHO 2004).
While AGI mortality is low in developed countries, AGI morbidity remains important.In the United States, children younger than 5 years of age are estimated to experience one to two episodes of diarrhea annually, with 300-400 deaths attributed to AGI each year (Glass et al. 1991).A study of approximately 300,000 American children younger than 5 years of age estimated that diarrhea accounted for 4% of all hospitalizations and 2% of all outpatient visits in this population.Based on the total outpatient and inpatient payments in this group, researchers estimated that the total national payment for diarrheaassociated disease in this age group was $411 million per year (in 1998 US dollars) (Zimmerman et al. 2001).This estimate would have been greater if the costs associated with morbidity and mortality in older children and adults were included.
Accurate estimates of AGI incidence and prevalence are essential for the development of sound public health policy.An overall estimate of the prevalence of AGI due to all etiologies is the necessary first step in developing an estimate of the prevalence of gastrointestinal illness due to specific exposures or etiologies, such as drinking water consumption.Disease surveillance plays an important role in developing these estimates.However, the data available from outbreak surveillance, laboratory-based communicable disease reporting, and other public health surveillance systems underestimate the rate of disease because they usually only capture case-patients in contact with the healthcare system.Not all persons infected by enteric pathogens develop gastrointestinal symptoms (Figueroa et al. 1983;Ish-Horowicz et al. 1989;Pettoello-Mantovani et al. 1995).Furthermore, less than a third of persons with an AGI seek medical care and stool samples for laboratory diagnoses are only obtained in a minority of these patients (Hawkins et al. 2002;Herikstad et al. 2002;Imhoff et al. 2004;Jones et al. in press).Most persons consulting a healthcare provider for a gastrointestinal illness are either treated with supportive care, regardless of the cause, or are treated presumptively without confirming the etiology through laboratory testing.Even if laboratory testing is ordered, different disease agents require different testing methodologies and laboratories may not perform the necessary diagnostic tests in every case (Jones et al. 2004;Voetsch et al. 2004).Furthermore, most stool tests fail to identify a pathogen (Magliani et al. 1985;Essers et al. 2000;Denno et al. 2005).While some cases of gastrointestinal illness due to laboratory-confirmed pathogens are required to be reported to public health agencies, most are not (CDC 2005a).Even when reporting is required, a large proportion of laboratoryconfirmed notifiable diseases go unreported (Standaert et al. 1995).This phenomenon results in underreporting of AGI, particularly of mild cases.Therefore, passive surveillance captures only a small fraction of the true rate of disease.
Because so many episodes of AGI are treated at home without contact with the healthcare system, special studies within communities are required to estimate the incidence and prevalence of gastrointestinal illness in the general population.In the United States, information on AGI incidence and prevalence has come from prospective community-based cohort studies, intervention trials, and retrospective population-based cross-sectional surveys.
Similar studies have been conducted in other developed countries.However, using the data from these studies to derive an estimate of AGI prevalence is still problematic.
The definition of what constitutes AGI is inconsistent between studies.Some studies only measure a single symptom -diarrhea -but even the definition of diarrhea varies.This paper reviews studies estimating the AGI incidence and prevalence in developed countries, examines the strengths and weaknesses of these studies, compares the estimated rates, discusses the limitations associated with such a comparison, and proposes an estimate of AGI prevalence in the United States based on available information.

METHODS
To identify the information available on the rate of AGI in developed countries, we conducted a literature review of peer-reviewed journal articles, scientific conference abstracts, and public health and environmental health agency documents available on the Internet.To identify the journal articles, we performed a PubMed search through the US National Library of Medicine using various combinations of the following key words: diarrhea, vomiting, gastroenteritis, gastrointestinal, prevalence, incidence, and human.Reference lists from relevant articles identified in the PubMed search were reviewed to find other relevant journal articles, abstracts, and agency reports.Using this approach, we identified 33 studies published since 1953 that provide estimates of AGI incidence and prevalence in the United States and in other developed countries.

RESULTS
Estimates from prospective community-based cohort studies and intervention trials in the United States Since the mid-1900s, there have been several large-scale prospective community-based cohort studies and intervention trials to determine the rate of AGI in the United States.
The first of these studies was conducted by Dingle et al. (1953Dingle et al. ( , 1964) ) from January 1 1948 to May 31 1957.
Investigators observed the occurrence of respiratory illness and unexplained gastrointestinal illness among families in Cleveland, Ohio over a 10-year study period.Families were selected in a non-random fashion through referral by family physicians or pediatricians.All had young children, belonged to the middle or upper socioeconomic classes, and were located within a specified residential area.Data was collected on the occurrence of diarrhea, vomiting, abdominal pain, and respiratory symptoms.One adult per family, usually the mother, kept a daily record of symptoms present in each family member.Field workers visited the homes weekly to check the records.During the early years of the study, staff physicians visited most of the ill study participants.During the later years of the study, persons with mild afebrile illnesses were not routinely examined.
In this study, illness was defined as the presence of one or more symptoms abnormal for that individual.Only gastrointestinal symptoms judged to be the result of primary gastrointestinal disease were considered in this 10-year study.Gastrointestinal symptoms thought to be secondary to another non-gastrointestinal illness or to some cause other than illness, and those thought to be too minor to warrant consideration as true symptoms were excluded.
Those illnesses determined to be primary gastroenteritis were further categorized as infectious or noninfectious.
The remaining gastrointestinal illnesses were assumed to be infectious in nature.An infectious gastroenteritis case was defined as two or more of the following symptoms: vomiting, abdominal pain, diarrhea, and fever (2139 cases).An infectious gastroenteritis-epidemiological case was defined as one of the above major symptoms (vomiting, abdominal pain, diarrhea, or fever) or minor symptoms (nausea, anorexia, etc.) occurring within 10 days of another infectious gastroenteritis case or infectious gastroenteritisepidemiological case in the same family (1169 cases).A probable infectious gastroenteritis case was defined as the same symptoms as an infectious gastroenteritis -epidemiological case but was not associated, within 10 days, with the onset of another case of infectious gastroenteritis (749 cases).In total, 439 members from 85 different families participated in the study and contributed 2692 person-years of follow-up.The median enrollment time per person was 2100 days (range 35 -3439 days).There were a total of 4057 reported episodes of infectious gastroenteritis for an incidence of 1.52 episodes per person-year.
A second study to examine gastroenteritis of unknown etiology was performed by the same investigators using the 1948-1950 subset of the data from the above study (Dingle et al. 1953(Dingle et al. , 1964;;Hodges et al. 1956).In this subset study, gastroenteritis was defined as any one or more of the following three conditions preceded by at least 5 symptom-free days: (1) diarrhea, (2) vomiting, or (3) abdominal pain.Participants reported 1466 episodes of gastroenteritis defined using these criteria.However, 362 episodes were considered to be secondary to other causes as explained by descriptions found in the records -these cases were not included in the analysis.Among these excluded cases were 116 episodes of gastrointestinal symptoms attributed to acute infectious diseases, such as streptococcal infections, measles, chickenpox, and bacterial gastroenteritis.Therefore, only 1104 episodes of infectious gastroenteritis of unknown etiology were analyzed for an overall incidence of 1.6 episodes per personyear (Hodges et al. 1956;Dingle et al. 1964).A large number of these unexplained episodes were hypothesized to represent acute, infectious, nonbacterial gastroenteritis.
Among the children in this cohort, infants younger than 1 year of age had the lowest incidence of infectious gastroenteritis of unknown etiology (1.2 episodes per person-year) and children 4 years of age had the highest incidence (2.6 episodes per person-year).Adults had an incidence of 1.2 episodes per person-year.The incidence of infectious gastroenteritis of unknown etiology was lowest in the summer from May to August and highest in the late autumn and early winter, peaking in October and November (Hodges et al. 1956).The proportion of the 1104 episodes associated with either diarrhea or vomiting cannot be determined from the information provided, but diarrhea occurred in 56% and vomiting in 53% of a subset of 683 cases of gastroenteritis that occurred at least 6 days after the onset of respiratory symptoms (Dingle et al. 1956(Dingle et al. , 1964)).Investigators observed that when gastrointestinal and respiratory symptoms in the same case-patient were temporally related, the onset of gastrointestinal symptoms was concurrent with or followed the onset of respiratory From November 1965 to August 1969, the Seattle Virus Watch Program followed a cohort of families to assess infections with respiratory or enteric viruses recoverable in cell cultures (Fox et al. 1972).Two groups of families were non-randomly recruited throughout the study period from a large comprehensive prepaid medical care plan.This plan covered approximately 10% of the population of the Seattle, WA metropolitan area.For logistical reasons, participants came from the northern two-thirds of metropolitan Seattle.
In total, 215 predominantly white middle-income families with newborn infants were enrolled for 2-year periods of observation.A nurse visited the first group of 149 families on a bi-weekly basis to collect and review illness records kept by the mother for the entire family.During these visits, routine fecal and respiratory specimens were collected.
Supplemental specimens were collected upon report of illness.Mothers were requested to notify investigators when illness occurred between visits.The 149 Group 1 families contributed 965 person-years of observation and approximately 85% of these families stayed in the study until their 2-year observation periods were completed or the study terminated.The 66 Group 2 families were either recruited after the monthly quota for Group 1 families was filled or preferred the less intense form of observation provided for Group 2. In Group 2, the mother was contacted weekly by telephone for information about illnesses in the family.Fecal and respiratory specimens were collected only upon report of illness and no more than twice a year.The 66 Group 2 families contributed 431 person-years of observation but 51% of these families withdrew from the study before their 2-year observation periods were completed or the study terminated.
This study used the same classification of enteric and respiratory illnesses as the New York Virus Watch study, although, again, the symptomatology of enteric illness was not defined.When the cases with enteric symptoms alone, without respiratory symptoms, were analyzed, the Group 2 families reported fewer enteric illness episodes than the  (Monto et al. 1971) and by 1968 these older families constituted approximately half of the participating households (Monto et al. 1970).One adult member of each household was contacted weekly, usually by telephone, and asked about the onset of acute illnesses among family members; illnesses were followed to their conclusion date during subsequent weekly telephone calls.
If at least 2 symptom-free days had passed between periods of reported illness, the latter illness was recorded as a new event (Monto et al. 1971) Insufficient data were provided by the authors to determine the separate incidence rates of each of the four enteric illness syndromes in the absence of respiratory symptoms.
Overall, 52% of enteric illness cases were accompanied by restriction of ordinary daily activity for an incidence of 0.63 activity-restricting enteric illness episodes per person-year.
The types of restriction of daily activity were not further defined.
From August 1975 to July 1977 investigators conducted a prospective community-based study of suburban families with one to four children in Charlottesville, VA (Hughes et al. 1978;Guerrant et al. 1990) (Payment et al. 1991(Payment et al. , 1997) ) and Australia (Hellard et al. 2001).They hypothesized that this was likely due to a loss of enthusiasm for the study.
Estimates from retrospective cross-sectional studies in the United States In addition to the prospective cohort studies and intervention Surveillance System (BRFSS) (Gentry et al. 1985;Remington et al. 1988;Tourangeau 2004;Jones et al. in press).
Telephone numbers in the FoodNet catchment area were screened using Genesys-ID software (Abt Associates Inc. 2002) to remove business and non-working numbers before contacting households using a single-stage random digit dialing sampling method (Jones et al. in press).One person per household was randomly selected to participate in the survey from a roster of household members of all ages.Parents or guardians responded for children younger than 12 years of age (Imhoff et al. 2004).Only English-speaking persons were interviewed until the fourth survey when the survey was also administered in Spanish (Jones et al. in press).
Respondents were asked about episodes of diarrhea in the four weeks (1996 -1997 and 1998-1999 survey cycles) or month (2000 -2001 and 2002-2003 survey cycles) prior to the interview (Jones et al. in press).Diarrhea was defined as three or more loose stools in a 24-hour period.Diarrheal illness was defined as diarrhea resulting in an impairment of daily activities (e.g.missing time from work, school, recreation or vacation activities, or work in the home) or diarrhea with a duration greater than 1 day.In the first and second cycles, respondents reporting diarrhea were asked about other symptoms experienced during their illness, including vomiting, fever, and abdominal pain.In the third and fourth cycles, all respondents, including those without diarrhea, were asked about vomiting; those respondents reporting any diarrhea or vomiting were also asked about concurrent respiratory symptoms.
A comparison of the prevalence of diarrheal illness from all four cycles of the FoodNet survey (1996 -1997, 1998 -1999, 2000 -2001, and 2002-2003)  For the purposes of this paper, we used the FoodNet population survey data to estimate the prevalence of AGI in the United States for the period 2000 -2003.As discussed, the first and second survey cycles did not record information on respiratory symptoms or vomiting without diarrhea.Therefore, only data from the third and fourth cycles were used for this estimate.We defined AGI as diarrheal illness and/or vomiting of infectious or noninfectious origin, excluding cases with concurrent respiratory symptoms (defined as cough and/or sore throat).
Overall, 8.8% of respondents reported diarrheal illness and/or vomiting in the month prior to interview; of these, 39% reported having respiratory symptoms.Therefore, the overall prevalence of self-reported AGI in the month prior to interview was 5.4%, a rate of 0.65 episodes per personyear.The prevalence of AGI did not differ significantly between the third and fourth cycles (Table 1).The prevalence of AGI was higher in females than males.
Children younger than 5 years of age reported the highest prevalence while persons 65 years of age and older reported the lowest prevalence.White respondents reported a higher prevalence than African Americans or Hispanics.The prevalence reported by adults with less than a high school education was lower than that of high school graduates.The prevalence of AGI was higher in the winter months compared to the summer months.et al. 2002;Hall et al. 2002Hall et al. , 2005;;Scallan et al. 2005).In this national gastroenteritis survey, private households across Australia with fixed telephone lines were sampled using random-digit dialing.The household member with the last birthday was selected to be interviewed.Interviews were conducted in seven languages.The incidence reported in this study was lower than that reported in the previous Dutch study.The investigators speculated that this decrease may have been due to an actual decrease over the decade but stated that it was more likely that the incidence in 1991 was overestimated because the response seemed to be strongly influenced by gastrointestinal symptoms.Other explanations could be that slightly different case definitions were used between the studies and that the earlier study period only covered part of a year and did not account for seasonality.Practice registries were used to sample study participants and enrollment was staggered over the 18 months of the study.Within each practice, persons were selected to be contacted using stratified random sampling by age and sex from the patient register.Two consecutive cohorts of patients were each followed for 6 months, rather than one cohort of 12 months, in order to increase the level of participation (Sethi et al. 1999).Participants returned weekly postcards for 6 months declaring the absence of intestinal illness.IID was defined as loose stools or significant vomiting (more than once in 24 hours, incapacitating, or accompanied by cramps or fever) lasting less than 2 weeks, in the absence of a known noninfectious cause and preceded by a 3-week symptom-free period (Sethi et al. 1999).Those who developed symptoms were asked to contact the study nurse, complete a risk factor questionnaire, and submit a stool sample.When recruited, each person was also asked to recall episodes of diarrhea in the month preceding recruitment to provide a retrospective estimate of the incidence of diarrhea studies requesting stool specimens, some still to be discussed, reported higher incidence rates (Fox et al. 1966(Fox et al. , 1972;;Hughes et al. 1978;Hoogenboom-Verdegaal et al. 1994;de Wit et al. 2000de Wit et al. , 2001)).The retrospective estimate of diarrhea incidence from this study was 0.55 episodes per person-year, which was nearly three times the prospective estimate.This finding has significant implications for comparisons made between retrospective and prospective studies of AGI, which will be discussed in greater detail in a subsequent section.

Norway
In

DISCUSSION
In this chapter we have reviewed 14 American and 19 international studies estimating the rate of AGI.The range of estimates provided varies both among and within countries.Some of this variation is due to differences in study methodology and design; this makes comparisons between studies problematic.
Of the 33 studies reviewed, 16 were retrospective studies and 18 were prospective studies (of which five were intervention trials with a prospective component and one was a recalculation of existing data).One study from England had both a retrospective and prospective component (Table 2).Both retrospective and prospective studies have some limitations.Retrospective studies suffer from a number of recall errors.One such error is telescoping.This phenomenon involves compression of time whereby an event is remembered as having occurred more recently than it actually did (Sudman & Bradburn 1973;Wheeler et al. 1999).The result is that events, such as the occurrence of AGI, may be over-reported, particularly if the event is severe, and thus, more memorable.This problem is more pronounced with proxy interviews for children (Bruijnzeels et al. 1998).At the same time, milder, less memorable symptoms may be underrepresented or even forgotten, particularly if the recall period is long.In the retrospective studies reviewed, the recall periods for AGI symptoms ranged from 2 weeks (Akhter et al. 1994) to 3 months (Palmer et al. 1996) but most were 1 month.Therefore, because of recall errors, retrospective studies may result in higher estimates of morbidity.Consequently, prospective studies are sometimes considered more reliable than retrospective studies because they reduce or eliminate recall errors.Wheeler et al. (1999) found that, within the same study, the retrospective estimate of infectious intestinal disease was almost three times higher than the prospective estimate.However, prospective studies also have difficulties.Four of the five intervention trials reviewed (Payment et al. 1991(Payment et al. , 1997;;Hellard et al. 2001;Colford et al. 2005) showed signs of reporting fatigue.Each of these trials required individual participation for over 1 year.
Over time, the rate of reported symptoms declined.Investigators hypothesized that participants lost interest in prospectively reporting their symptoms, which may have resulted in an underestimation of the disease rate.Furthermore, some prospective studies collected stool samples when symptoms were reported.Participants in these studies may have been unwilling to report symptoms because of the follow-up such a report triggered, once again leading to underreporting.Given the types of errors associated with retrospective and prospective studies, one would anticipate that retrospective studies would, in general, provide higher estimates of AGI rates than prospective studies.This is true in a comparison of Australia, British, and Canadian studies in which all estimates from retrospective studies were higher than those provided by the prospective studies in the same country.However, this trend does not appear to be true in the American studies.In the United States, the range of estimated AGI rates from retrospective and prospective studies overlapped: 0.6 episodes per person-year (Hawkins et al. 2002;Jones et al. in press) to 3.2 (Sandler et al. 2000) in the retrospective studies, and 0.3 (Fox et al. 1972) to 3.48 (Colford et al. 2002) in the prospective studies.The cause of this discrepancy is unclear but may be related to the use of different case definitions in the different countries.These case definitions used in the retrospective Australian, Canadian, and British studies tended to be similar to those used in the prospective studies from the same countries.
These case definitions often referred to both diarrhea and vomiting.In contrast, all of the retrospective American studies used a restrictive definition of AGI that included diarrhea only, whereas the prospective American studies used broader definitions of AGI that included diarrhea or vomiting, with many also including a combination of other symptoms.The two prospective studies by Fox et al. (1966Fox et al. ( , 1972) ) did not provide detailed case definitions so comparisons using these studies are not possible.The use of restrictive case definitions in the retrospective American  2), even the definitions of the symptoms themselves vary.For example, diarrhea was defined as loose stools or stools with abnormal liquidity (Majowicz et al. 2004), loose stools present for fewer than 14 days (Roderick et al. 1995), two or more loose stools a day (Hoogenboom-Verdegaal et al. 1994;de Wit et al. 2000;Hellard et al. 2001), or liquid versus soft stools (Payment et al. 1991(Payment et al. , 1997)).tinal illness and found that 99% of the combined study population had bowel movements ranging from three per week to three per day.More recent studies have also found similar rates in adults (Drossman et al. 1982;Bassotti et al. 2004) and children 1 -4 years of age (Weaver & Steiner 1984).Connell et al. (1965) concluded that more than three bowel movements per day might be considered unusual.
They also observed a correlation between increasing stool frequency and the person's opinion of the stool as loose.
However, limited information is available about the validity of self-reported diarrhea where specific symptoms are not defined.Baqui et al. (1991) tested various definitions of diarrhea in Bangladeshi children younger than 5 years of age using prospective community-based surveillance data and found that reports of either (1) three or more loose stools in a 24-hour period or (2) any number of loose stools containing blood in a 24-hour period seemed to be the most sensitive (78%) and specific (96%) definition compared to the mothers' perceptions of diarrhea.Sandler & Drossman (1987) studied young adult university students and new hospital employees and asked them to define diarrhea.Most (84%) included loose or watery stools in their definitions while only about a quarter of participants included urgency (27%), frequent stools (26%), and abdominal discomfort (24%).These definitions were not mutually exclusive.Talley et al. (1994) studied an age-and gender-stratified random sample of residents 20-64 years of age in Olmsted County, Minnesota and found that self-reported diarrhea identified only 39% of the participants who reported one or more of four major diarrheal symptoms: (1) loose or watery stools more than 25% of the time; (2) a stool frequency often of more than three per day; (3) a stool frequency usually of more than 21 per week; and (4) urgency.Investigators found that the overlap between self-reported diarrhea and the presence of individual symptoms was greatest for loose or watery stools and urgency, with estimates of stool frequency of lesser importance.They also found that diarrheal symptoms inadequately discriminated between self-reported diarrhea and self-reported normal bowel habit, thereby raising the philosophical question about what constitutes diarrhea and whether the definition of diarrhea should encompass some measure of a change from normal bowel habit or, as a surrogate, a measure of severity (such as the duration of diarrhea or the effect of diarrhea on behavior).These studies indicate that self-reported diarrhea alone, without clarification of the symptoms, is not an adequate measure of diarrheal illness and should be used with caution in clinical trials and epidemiological studies.
The study by Sandler et al. (2000), previously discussed, that provided the outlying estimate of AGI prevalence might be an example of this phenomenon.Alternatively, specification of loose or watery stools and the frequency of stools (i.e. three or more in a 24-hour period) in the definition of diarrhea may improve the clinical and epidemiologic validity of the response.Some studies have attempted to address these issues by defining diarrhea and incorporating a measure of severity into their definitions of AGI.For example, the four FoodNet survey cycles (Herikstad et al. 2002;Hawkins et al. 2002;Imhoff et al. 2004;Jones et al. in press) define diarrheal illness as diarrhea (three or more loose stools in a 24-hour period) resulting in an impairment of daily activities (e.g.missing time from work, school, recreation or vacation activities, or work in the home) or diarrhea duration greater than 1 day.Such a definition addresses the departure from normal bowel habits (of which there is a range) and identifies more severe cases that have a personal impact.However, by using a more specific definition of AGI, the sensitivity for mild cases is reduced.
Therefore, the balance between sensitivity and specificity of the definition of AGI needs to be determined by the purpose of the investigation in which the definition is being used.
The definition of an episode of AGI not only requires specification of the symptoms involved, it also requires identification of the end of an illness episode.One study has found that 3 intervening diarrhea-free days seemed to be the optimal interval to define new episodes of diarrheal illness (Baqui et al. 1991).This finding is supported by Morris et al. (1994), who modeled the distribution of illness episodes and found that an interval of 2 or 3 days without symptoms generally marked a new episode of diarrhea.However, the intervals used in the studies reviewed in this paper ranged from 2 days (Monto & Koopman 1980) to 3 weeks (Roderick et al. 1995;Wheeler et al. 1999), with a median of 6 days.Therefore, some of these studies may have misclassified two or more distinct AGI episodes as a continuation of a single episode.Misclassification was not the only mechanism by which underreporting of AGI may have occurred.All the retrospective studies counted only one episode of AGI during the period of interest, even if more than one episode was experienced.Regardless of the definition used, those persons with mild symptoms that did not meet the case definitions were not counted.Therefore, all but the very broadest case definitions may have underestimated the rate of mild cases.
Another way in which the 33 studies reviewed varied was in the method of contact with study participants.Four studies (Hodges et al. 1956;Fox et al. 1966;Hughes et al. 1978;Feldman & Banatvala 1994) conducted in-person interviews to gather information on symptoms and illness.
Two studies (Fox et al. 1972;Payment et al. 1991) used a combination of methods, and the authors of one study (Strauss et al. 2001) did not specify the method of participant contact.These differing methodologies are subject to different response rates and nonresponse biases.
Studies have found that in-person interviews tend to have the highest response rates, followed by mail surveys, then telephone surveys (Marcus & Crane 1986;Picavet 2001).
Only one of the in-person studies reviewed supplied a response rate (77%) (Feldman & Banatvala 1994).The response rates for nine of the mail studies ranged from 31% (Roderick et al. 1995) to 96% (Colford et al. 2002).The response rates for eight of the telephone studies were comparable to those by mail.Of note, the FoodNet population telephone survey response rate declined over the four survey cycles, from 71% in the first cycle (Herikstad et al. 2002) to 33% in the fourth cycle (Jones et al. in press).This is representative of an overall decline in survey response rates (Atrostic et al. 2001;Tourangeau 2004).In particular, telephone surveys in the United States have been affected by the increase in private telemarketing, and the introduction of "do-not-call" lists and caller screening devices (Tourangeau 2004).The growth in households using only cellular telephones (without a land line) may also be of concern.During the first half of 2005, CDC's National Health Interview Survey found that 6.7% of adults had access only to cellular telephones.Cellular-only usage was more common among certain groups, such as young adults and persons renting their homes.However, investigators concluded that, while the percent of adults without land line telephones has increased, it is still low, which minimizes the bias resulting from their exclusion from telephone surveys (Blumberg et al., 2006).While several recent studies have demonstrated little relationship between nonresponse rates and nonresponse bias (Curtin et al. 2000;Keeter et al. 2000;Tourangeau 2004), low response rates are still problematic when trying to generalize results to both the population under study and to a wider population.Inperson interviews, mail surveys, and telephone surveys are also each subject to different nonresponse biases, whereby the group of people not interviewed may be systematically different from those that are interviewed.Compared to inperson interviews, mail surveys are less likely to be completed by persons with lower levels of education and literacy (Picavet 2001).Telephone surveys are also less likely to reach low-income minorities and persons with lower educational levels (Marcus & Crane 1986;Imhoff et al. 2004).All three methods exclude institutionalized persons (e.g.persons in long-term healthcare facilities, mental institutions, and jail), persons who could not respond because of physical or mental impairment, and persons speaking languages that are different than the language(s) of the interview or survey.If these excluded persons are different from the larger population concerning the event of interest (i.e.AGI), then bias may be introduced and estimates of AGI extrapolated to the US population may be inaccurate (Imhoff et al. 2004).
Selection bias is another means by which persons under study differ from the rest of the population, thereby limiting the generalizability of the results.This bias may be present at the level of the respondent.Persons with AGI may be more likely to complete a survey or interview, thereby creating a selection bias.Selection bias may also be present within the sampling frame.Most of the studies reviewed in this chapter evaluated relatively small specific groups of people.Eight studies (Hodges et al. 1956;Fox et al. 1966Fox et al. , 1972;;Monto & Koopman 1980;Guerrant et al. 1990;Payment et al. 1991Payment et al. , 1997;;Hellard et al. 2001) were limited to families with young children.Therefore, children were overrepresented in these studies, which could have inflated the estimated AGI rate because children are known to experience higher rates of AGI than adults (Hodges et al. 1956;Monto & Koopman 1980;Hawkins et al. 2002;Herikstad et al. 2002;Imhoff et al. 2004;Jones et al. in press).Five studies were conducted in populations served by specific medical practices or health insurance organizations (Fox et al. 1972;Roderick et al. 1995;Palmer et al. 1996;Wheeler et al. 1999;de Wit et al. 2001), eliminating the medically indigent and those outside the service areas.
Only four international studies (Hoogenboom-Verdegaal et al. 1994;Hall et al. 2002;Kuusi et al. 2003;Scallan et al. 2004) assessed the national rate of AGI using a representative sample of the nationwide population.In the United States, Sandler et al. (2000) conducted a nationwide survey but, as just mentioned, only a small proportion of participants were randomly selected.The four cycles of the US FoodNet population survey (Hawkins et al. 2002;Herikstad et al. 2002;Imhoff et al. 2004;Jones et al. in press) were population-based but the FoodNet sites were not chosen to be representative of the US general population.
Rather, these sites were chosen based on their ability to conduct population-based surveillance and to achieve geographic diversity within their areas (Hardnett et al. 2004).However, this geographic diversity presents a problem.Studies of different enteric pathogens have demonstrated regional differences in the incidence of specific laboratory-confirmed infections (Hedberg et al. 1997;Bender et al. 2004;Ray et al. 2004) and have suggested that these variations in incidence reflect regional differences in physician and laboratory practices and perhaps regional differences in the risk of exposure (Hedberg et al. 1997;Hardnett et al. 2004).FoodNet has conducted surveys of physician and laboratory practices in FoodNet catchment areas and no differences were observed between FoodNet sites.However, because true regional differences in infection rates appear to exist based on other studies, these regional differences may have impacted the crude rate of AGI as estimated by the FoodNet population surveys as new sites were included in subsequent survey cycles (Hardnett et al. 2004).Demographic differences between the FoodNet and US populations must also be considered if generalizations using these data are to be made.In 1996, investigators compared the FoodNet and US populations and found that the age and sex distributions were similar but that the FoodNet population overrepresented Asians, underrepresented Hispanics, had a lower population density in FoodNet counties, and had a smaller percentage of persons living at or below the poverty level (Hardnett et al. 2002(Hardnett et al. , 2004)).A similar comparison of the FoodNet and US populations in 2000 again found similar age and sex distributions but, this time, a similar proportion of Asians in both populations, although Hispanics were still underrepresented in the FoodNet population (6% in FoodNet areas in 2000 compared to 12% nationally) (Hardnett et al. 2004).Overall, FoodNet researchers believe that the demographic differences appear to be limited.Furthermore, after accounting for the changing composition of the FoodNet sites between cycles by weighting for age, sex, location, and number of residential telephone lines, the estimates of diarrheal illness are comparable across the four survey cycles.Therefore, the FoodNet population survey data have been generalized to the US population.However, whenever this is done, it is important that the limitations of the data be well understood.
Seasonality of AGI is yet another issue to consider when comparing these studies or generalizing their findings.
Studies conducted over periods of less than one year (Akhter et al. 1994;Feldman & Banatvala 1994;Hoogenboom-Verdegaal et al. 1994;Roderick et al. 1995;Palmer et al. 1996;Sandler et al. 2000;Strauss et al. 2001;Ashbolt et al. 2002;Colford et al. 2002;Queensland OzFoodNet 2002;Gofti-Laroche et al. 2003) fail to capture the seasonal variation of AGI and, therefore, may overestimate or underestimate the annual rate of disease, depending on what time of year the study was conducted.
In temperate climates, gastrointestinal illness is reported to have a bimodal distribution.Bacterial gastroenteritis tends to peak in the summer months (Gurwith & Williams 1977;Michel et al. 1999;Denno et al. 2005) while viral gastroenteritis, which may be more common (Gurwith & Williams 1977), tends to peak during the winter months (Gurwith & Williams 1977;Cook et al. 1990;Mounts et al. 2000;Denno et al. 2005) with lower rates in the summer (Hodges et al. 1956;Monto & Koopman 1980;Payment et al. 1997;Kuusi et al. 2003;Jones et al. in press).
Another barrier to comparability is that different studies used different exclusion criteria for individual participants or cases.Therefore, the estimated AGI rates were calculated from different groups of people, making comparisons problematic.Eight studies (Hughes et al. 1978;Akhter et al. 1994;Feldman & Banatvala 1994;Palmer et al. 1996;Hoogenboom-Verdegaal et al. 1994;Colford et al. 2002Colford et al. , 2005;;Gofti-Laroche et al. 2003) did not report any exclusion criteria for individual participants or cases.Eleven studies (Hellard et al. 2001;Ashbolt et al. 2002;Hall et al. 2002;Herikstad et al. 2002;Kuusi et al. 2003;Imhoff et al. 2004;Hawkins et al. 2002;Queensland OzFoodNet 2002;Majowicz et al. 2004;Scallan et al. 2005;Jones et al. in press) excluded persons reporting a chronic illness in which diarrhea was a major symptom.Therefore, their estimated rates failed to account for AGI episodes in chronically ill persons that were unrelated to their chronic illnesses or conditions.The first two FoodNet surveys (Herikstad et al. 2002;Imhoff et al. 2004) also excluded persons who had surgery to remove parts of their stomachs or intestines because these surgeries predisposed them to recurrent noninfectious diarrhea.However, this meant that AGI episodes unrelated to surgery were not counted.The Cleveland study (Dingle et al. 1953(Dingle et al. , 1964;;Hodges et al. 1956) excluded cases of AGI with known etiologies, some of which were bacterial gastroenteritis.Therefore, this study underestimated the rate of infectious gastroenteritis.Six studies (Payment et al. 1991(Payment et al. , 1997;;Roderick et al. 1995;Wheeler et al. 1999;de Wit et al. 2001;Scallan et al. 2004) excluded noninfectious causes.Therefore, their estimated rates of AGI were limited to infectious etiologies and may have excluded environmental causes, such as chemical exposure.
Five studies (Hodges et al. 1956;Fox et al. 1966Fox et al. , 1972;;Monto & Koopman 1980;Garthright et al. 1988) provided estimates of AGI rates in the absence of respiratory symptoms.The Cleveland study found that gastrointestinal illness was associated with respiratory symptoms in 20% of cases (Hodges et al. 1956;McCorkle et al. 1956), often with gastrointestinal symptoms beginning at the same time as respiratory symptoms or shortly thereafter (McCorkle et al. 1956).In contrast, respiratory illness was associated with gastrointestinal symptoms in 5% (McCorkle et al. 1956) of cases.In the Cleveland study, respiratory illness included common respiratory diseases (e.g. common cold, rhinitis, laryngitis, bronchitis, and other acute respiratory illnesses of undifferentiated type) and specific respiratory diseases (e.g.streptococcal tonsillitis and pharyngitis, nonstreptococcal exudative tonsillitis and pharyngitis, primary atypical pneumonia, pneumococcal pneumonia, and influenza) (Dingle et al. 1953).In the Tecumseh study (Monto & Koopman 1980), gastrointestinal illness was associated with respiratory symptoms in 27% of cases and respiratory illness was associated with gastrointestinal symptoms in 11% of cases.In this study, respiratory illness was divided into five syndromes: (1) lower respiratory illness with a productive cough, wheezing, or pain on respiration, (2) upper respiratory illness with coryza, without lower respiratory symptoms, (3) laryngotracheal illness with sore throat or hoarseness, without lower or upper respiratory symptoms, (4) nonproductive cough, and (5) earache alone (Monto & Koopman 1980).In the Cleveland and Tecumseh studies, many cases with combined gastrointestinal and respiratory symptoms appeared to have a common etiology.However, because it was unclear as to whether these were primarily gastrointestinal or respiratory illnesses, cases with combined symptoms were excluded from these analyses.
Recent studies have attempted to exclude persons with gastrointestinal symptoms secondary to respiratory infections to more accurately reflect the true rate of AGI (Mead et al. 1999;Hall et al. 2005) All of these methodological variations make comparisons between the studies difficult and interpretations of the estimated AGI rates should be made with caution.Rates of AGI differ within and between study types, and within and among countries.Taken as a whole, these 33 studies suggest that the rate of AGI (including diarrhea) in developed countries is somewhere in the range of 0.1 (Roderick et al. 1995) to 3.5 (Colford et al. 2002) episodes per person-year, depending on location and type of study.Among the international studies, the range is 0.1 (Roderick et al. 1995) to 2.8 (Gofti-Laroche et al. 2003).In the United States, the same range is 0.3 (Fox et al. 1972) to 3.5 (Colford et al. 2002).These estimates come from a variety of different study types, including retrospective cross-sectional population-based surveys, prospective cohort studies, and intervention trials designed to assess the rate of diarrhea and AGI.Other study types peripherally capturing information on AGI can also inform estimates of the rate of AGI.For example, a convenience sample of three randomized, double-blind, placebo-controlled drug trials in the US (Tilley et al. 1995;Black et al. 1997;Szapary et al. 2003) reported a range in the rates of diarrhea among their placebo groups (adults with active rheumatoid arthritis, essential hypertension, and primary hypercholesterolemia, respectively) of 0.05 (hypertensive placebo group) to 0.87 episodes per person-year (hypercholesterolemic placebo group).This range overlaps with the ranges presented in the studies specifically assessing AGI.

CONCLUSIONS
Within the limitations described previously, the FoodNet studies are the most generalizable to the US population given their study design.They likely provide the best data currently available for an estimate of the rate of AGI in the United States.Their retrospective study designs could have resulted in over-reporting and may have lead to an overestimate of the rate of AGI.Using the data from the third and fourth FoodNet survey cycles, the estimated rate of AGI in the U.S. is 0.65 episodes per person-year, with an unknown degree of uncertainty around this point estimate.However, this estimate does fall within the range of estimates presented by other national and international studies of varying designs.
For this FoodNet estimate, AGI was defined as diarrheal illness (three or more loose stools in a 24-hour period resulting in an impairment of daily activities or diarrhea duration greater than 1 day) and/or vomiting, excluding those with respiratory symptoms (cough and/or sore throat).
The diarrhea and/or vomiting could have been of either infectious or non-infectious origin.However, this definition excluded episodes of diarrhea or vomiting due to any longlasting or chronic illness or condition.This case-definition of AGI is supported by studies in the literature that indicate that the validity of self-reported diarrhea is improved by including "three or more stools in 24 hours" and "loose stools" in the case definition.A measure of severity was added to ensure mild, noninfectious causes of AGI were excluded.Vomiting was included in the case definition to estimate the rate of AGI rather than just the rate of diarrheal illness.Other AGI symptoms (e.g.nausea, abdominal pain) were not considered sufficient to meet the case definition in the absence of diarrhea or vomiting.Finally, cases with gastrointestinal symptoms secondary to respiratory illnesses were excluded to improve the specificity of AGI rate estimates.We believe that this case definition and the estimate of 0.65 AGI episodes per person-year can serve as a basis for the calculation of the rate of endemic gastrointestinal illness due to public drinking water systems in the United States.
symptoms; gastrointestinal symptoms did not precede respiratory symptoms.Of the 1104 unexplained gastrointestinal illness cases analyzed, approximately 20% comthe gastrointestinal and respiratory symptoms presumably had a common etiology and in which gastrointestinal symptoms began the same day as or within a few days after respiratory symptoms.Similarly, this syndrome accounted for an estimated 5% of common respiratory disease in this population(McCorkle et al. 1956).When persons with respiratory -gastrointestinal syndrome were excluded, the incidence of infectious gastroenteritis of unknown etiology was 1.3 episodes per person-year.This figure would likely have been slightly higher if some of the cases of known bacterial gastroenteritis had been included in the analysis.From August 1 1961 to March 31 1965, the New York Virus Watch Program followed a highly select population in the metropolitan New York area to gain information concerning viral agents detectable by in vitro culture methods available at that time(Fox et al. 1966).Two contrasting communities were selected for the study: a large urban housing project and a relatively isolated rural island community.Both were middle-income, predominantly white communities.Investigators used non-random selection techniques to enroll families with children 1 -10 years of age.Families with newborns were recruited partway through the study.In each family, a parent recorded any illness in the family on a form that a nurse would pick up and review on a bi-weekly basis.At this time, routine fecal and respiratory samples were collected.Specimens were also collected upon report of illness.Signs and symptoms were coded by the field nurse and later reviewed by a pediatrician or a supervising nurse.Illnesses were then classified into categories based on the predominant symptoms.In total, 178 families with 791 persons were enrolled and contributed 882 person-years of observation.The investigators analyzed the occurrence of enteric and respiratory illness, though they did not provide a definition for enteric illness in their paper.There were approximately 844 episodes of enteric illness for a rate of 1.0 episode per person-year.When cases with combined respiratory and gastrointestinal symptoms were excluded, there remained approximately 630 cases of enteric illness alone for a rate of 0.7 episodes per person year.
is awaiting publication(Jones et al. in press).For this comparison, investigators equated the 28-day period used in the first two cycles with the 1-month period used in the last two cycles.If more than one diarrheal episode occurred during the period of interest, only the most recent episode was considered.Respondents who indicated they had any long-lasting or chronic illness or condition in which diarrhea or vomiting was a major symptom (such as irritable bowel syndrome, ulcerative colitis, partial removal of stomach or intestines, stomach or esophagus problems, or Crohn's disease) were excluded from the analysis.Persons reporting concurrent diarrhea and respiratory symptoms were not excluded.Data from each cycle were weighted by age, sex, and location, using the projected census numbers from the corresponding years, to ensure that the survey populations were demographically representative of the FoodNet catchment areas.Data were also weighted using the number of eligible respondents per household and the number of telephone lines per household to compensate for the unequal probabilities of selection between and within households.Cooperation rates were calculated using the upper bound response rate formula provided by the Council of American Survey Research Organizations (CASRO)(White 1983;Remington et al. 1988).These rates include survey refusals, terminations, and completed interviews.The upper bound CASRO response rate declined over time from 71% in the 1996 -1997 survey cycle to 33% in the 2002 -2003 survey cycle(Jones et al. in press).
During the late 1990s,Hellard et al. (2001) conducted a drinking water intervention study in Australia.In this randomized, double-blinded controlled trial, households in Melbourne that (1) received their water from a defined catchment area; (2) had at least four eligible family members, including at least two children aged 1 -15 years; and (3) owned or would be purchasing their homes were eligible for the study.In addition to the household eligibility criteria, each participant had to consume at least one glass of tap water per day.Individuals who had immunosuppression, had a chronic diarrheal illness, or were on long-term antibiotic therapy were excluded.Invitations to participate were distributed by mail, through advertisements in local newspapers, and through primary schools, child-care centers, maternal health centers, and shopping malls.Interested families were asked to phone the study center if they met the inclusion criteria.Eligible families who called the study center were then mailed an information booklet about the study, visited at home, and enrolled.In total, 600 families (2811 individuals) were recruited and were randomly assigned to receive real or sham water treatment units installed in their kitchens.Families were blinded as to the type of unit they received.Highly credible gastroenteritis (HCG) was defined as any of the following symptoms in a 24-hour period: (1) two or more loose stools; (2) two or more episodes of vomiting; (3) one loose stool with abdominal pain or nausea or vomiting; or (4) one episode of vomiting with abdominal pain or nausea.Cases were considered distinct if the participant experienced at least 6 symptom-free days in between episodes.Families completed and returned health diaries by mail every 4 weeks.Participants were asked to collect fecal specimens during any episodes of HCG.A total of 600 families (2811 persons) were enrolled in the study and contributed 3333 person-years of health diary data during a 68-week period between September 1997 and February 1999 (excluding two 4-week periods over each Christmas season).There were 1352 cases of HCG reported during this period among persons with the sham units, for an incidence of 0.82 cases per person-year.Investigators noticed a decline in the reported incidence of HCG from the first 13 weeks of the study compared to the last 13 weeks.They believed this reflected an underreporting of symptoms due to declining motivation over time.Investigators found no significant difference in the incidence of HCG between the different treatment groups.During a 12-month period from September 2001 to August 2002, the National Centre for Epidemiology and Population Health, on behalf of OzFoodNet, conducted a national cross-sectional computer-assisted telephone interview (CATI) survey to estimate the incidence of infectious gastroenteritis (Ashbolt Four years later,Payment et al. (1997) conducted another intervention trial in the same suburban Montreal community.This time households were randomly chosen from a list of families in the study area who were enrolled in a government income supplement program for families with children younger than 18 years of age (this program was independent of income level).Households were eligible if they (1) were located within the distribution system of the water filtration plant; (2) had French-speaking occupants; (3) had occupants who regularly consumed tap water; (4) had at least one child between the ages of 2 -12 years living in the household; and(5) were willing to participate.For 16 months (September 1993 to December 1994), families with young children were randomly assigned to one of four groups: (1) regular tap water; (2) tap water from a continuously purged tap valve; (3) bottled plant water; and (4) purified bottled water.Participants were not blinded to their treatment status.A family diary of gastrointestinal and respiratory symptoms was kept and families were contacted by telephone every 2 weeks to obtain the information.The definition for an episode of highly credible gastrointestinal (HCGI) illness was the same as that used in the previous study.In total, 1062 families (5253 individuals) participated in the trial.The incidence of HCGI illness among the regular tap water group was 0.66 episodes per person-year.As with the previous study, this rate was adjusted using Poisson regression methods to account for the correlation between repeated episodes in the same person.The incidence of HCGI illness was highest among children younger than 6 years of age and lowest among adults 50 years of age or older.The highest incidence of HCGI illness was observed during autumn and winter and the lowest during the summer.The incidence was highest at the beginning of the trial in September 1993 and dropped steadily as the trial progressed.A similar trend was seen in the Davenport(Colford et al. 2005) and earlier Montreal(Payment et al. 1991) trials, as well as in an Australian study(Hellard et al. 2001).However, owing to a 16-month study period, a rising trend beginning in September 1994 was observed, suggesting that the seasonal increase in the winter months may have been present in spite of what appeared to be reporting fatigue.From February 1994 to February 1995,Raina et al. (1999) conducted a prospective cohort study of rural families drinking untreated well water to determine the relationship between consumption of E. coli-contaminated well water and gastrointestinal illness.Families from southern Ontario were recruited from participants in the Ontario Farm Groundwater QualitySurvey (1991Survey (  -1992) )  (Goss et al. 1998).This earlier study sampled water from 1292 of the estimated 500 000 water wells in Ontario, and the study conformed to a stratified random survey.It is unclear how the families participating in the Raina study were selected from this earlier study.Families in the Raina study were excluded if(1) they no longer used the same well as in the 1991 -1992 study; (2) the household would contain fewer than two fulltime residents during the study period; (3) the family did not drink well water; or (4) the family routinely treated the water.Children younger than 1 year of age at the start of the study and any persons absent from the household for more than 2 months during the study period were excluded from analysis.One contact person in each household completed a health diary for each family member.Interviewers telephoned the contact persons approximately once per month to collect the diary information.Gastrointestinal illness episodes were defined as diarrhea, with or without vomiting, occurring for 1 or more days with at least 5 symptom-free days separating episodes.The definition of diarrhea used in this study was not provided.Of the 442 families identified for potential inclusion in the study, 156 (35%) were enrolled and included in the final analysis.These 156 families included 531 individuals.Of these, 414 persons were considered to have non-contaminated water supplies during the study period because their wells tested negative for E. coli in all of five separate water quality tests staggered throughout the year.Among these 414 persons with non-contaminated well water, 25.8% reported at least one episode of gastrointestinal illness during the year, for an incidence of at least 0.26 episodes per person-year (counting only one episode per person per year).Gastrointestinal illness episodes were most frequent in February and March and fluctuated throughout the study period.For the 3-month period from April 3 1995 to July 22 1995, Strauss et al. (2001) conducted a study to examine the relationship between microbiologic contamination of drinking water from private wells in rural communities in easternOntario and the incidence of AGI.Four rural communities were selected, representing a cross section of rural population in the area.The selection criteria for the communities were not described.Households within these communities were randomly selected using a phone book database.All households within each community were eligible for participation except those not consuming drinking water from a private well, and residents of retirement or nursing homes.Health information was collected on each household member using self-reported health diaries for a 28-day period.AGI episodes were defined as either (1) vomiting or liquid diarrhea, or (2) nausea or soft, loose diarrhea with abdominal cramps.Episodes included 1 or more symptomatic days, with at least 6 consecutive symptom-free days between episodes.For analytic purposes, only the first episode was considered.Of the 327 households initially contacted, 235 (72%) representing 647 persons agreed to participate.Of these persons, 619 (96%) completed the diaries for the full 28 days and were included in the analysis.One or more AGI episodes were identified in 51 (8.2%) participants for an annualized incidence of 1.1 episodes per person-year.Seasonality was not accounted for in this 3-month study period.Children 10 years of age or younger were found to be more likely to have AGI symptoms.From February 2001 to February 2002, Majowicz et al. (2004) conducted a retrospective cross-sectional telephone survey in the city of Hamilton to assess the magnitude and distribution of self-reported AGI.Participants were randomly selected from a list of Hamilton residential telephone numbers in an electronic directory.One individual per household was randomly selected to participate by identifying the person with the next birthday.Interviews were conducted over 12 months, with approximately the same number of interviews completed each month.AGI was defined as any vomiting or diarrhea (loose stool or stool with abnormal liquidity) in the 28 days prior to the interview.Episodes due to other plausible etiologies (e.g.excess alcohol intake) were not excluded.Episodes due to pre-existing conditions (e.g.Crohn's disease, irritable bowel syndrome, lactose intolerance, and pregnancy) were excluded from the numerator but not the denominator.Incidence rates were adjusted to account for the likely proportion of pre-existing cases that actually began before the 28-day period.Of the 9543 persons contacted to participate in the study, 3496 (37%) agreed to participate.Of these, 351 reported an episode of AGI in the preceding 28 days; 138 reported more than one episode; however only the last episode was counted to minimize the potential for recall bias.The adjusted incidence was 1.3 episodes per person-year.The incidence of AGI was highest among children 0 -9 years of age (16%) and adults 20-24 years of age (18%) and lowest among adults 70 -74 years of age (4%).The incidence peaked in April, with another smaller peak in October.FranceBetween October 1998 and June 1999, Gofti-Laroche et al. (2003) conducted a prospective cohort study in the French Alps as part of a larger study to assess the risks of acute digestive conditions (ADC) in relation to protozoal contamination of drinking water.Volunteer families in communities in the Ise `re and Savoie departments of southeast France that were supplied by four public drinking water systems were recruited through notices distributed through the media, schools, and town councils, and from drinking water utility files.All four water systems were considered vulnerable to microbial contamination.Each family completed self-administered daily questionnaires concerning the health problems of each family member.Each weekday, 20% of the families were telephoned by an interviewer who recorded all incident cases of ADC occurring since the previous call.An ADC was defined as an episode of abdominal pain, nausea, vomiting, and/or diarrhea.The definition of diarrhea used in this study was not provided.A diarrheic episode (DE) was defined as diarrhea with at least one other digestive condition (unspecified) or fever.A case of gastroenteritis (GE) was defined as an episode of diarrhea with fever or vomiting.A total of 176 households, representing 544 persons, were enrolled in the study providing 252.6 person-years of observation.These persons were not representative of the local population with respect to socioeconomic status (white-collar workers and employees were over-represented; blue-collar workers and farmers were under-represented) and age (children were overrepresented).There were 712 reported cases of ADC, 105 cases of DE, and 46 cases of GE among all 4 drinking water systems combined.The overall annualized incidence rates per person-year were 2.8 for ADC, 0.4 for DE, and 0.2 for GE.The Netherlands Between March and July 1991, a population-based prospective cohort study was conducted in four regions of the Netherlands (Hoogenboom-Verdegaal et al. 1994; de Wit et al. 2000).Ten municipalities were chosen to represent the different geographical regions across the country: three rural, two urban, and five mixed rural/urban.Participants were randomly selected from the population registers of each municipality; only one person per household was eligible for the study.Participants were asked to return weekly questionnaires about the presence and duration of gastroenteritis.Two grades of gastroenteritis were defined.Grade 1 constituted diarrhea (two or more stools a day) or vomiting with two or more additional symptoms occurring within the 1-week period.Grade 2 constituted diarrhea or vomiting with two or more additional symptoms occurring on the same day and lasting at least 2 days within the 1week period.The additional symptoms included nausea, abdominal pain, cramps, and blood or mucus in the stool.Grade 2 gastroenteritis was more severe and defined a subset of Grade 1 gastroenteritis.An episode of gastroenteritis was considered to be over if 2 weeks had passed without further symptoms.Participants were also asked to submit a stool sample if symptoms developed.Of the 6243 persons invited, 2257 (36%) agreed to participate.Over the 17-week study period, there were 425 Grade 1 cases (0.57 cases per person-year), of which 115 (0.15 cases per personyear) could also be classified as Grade 2. The annualized incidence of acute gastroenteritis was highest among children younger than 10 years of age (0.98 cases per person-year) and lowest among persons 60 years of age or older (0.30 cases per person-year) (de Wit et al. 2000).Several years later, de Wit et al. (2000) recalculated the annualized incidence of Grade 1 gastroenteritis standardizing for age and sex using the 1991 Dutch mid-year population and estimated it at 0.45 cases per person-year.Investigators stated that the standardized incidence was lower because of an over-representation of children under 10 years of age.In the late 1990s, de Wit et al. (2001) conducted a prospective medical practice-based cohort study (called Sensor) to assess the incidence of gastroenteritis in the Netherlands.To include an entire year of data in the study, the investigators used two consecutive 6-month cohorts from December 14 1998 to June 13 1999 and from June 14 1999 to December 13 1999.Participants were recruited by age-stratified random sampling of all persons registered at 44 general practices within a sentinel network.Participants were asked to return a history card every week reporting the presence or absence of gastrointestinal symptoms in the previous 7 days.Those who developed diarrhea or vomiting were also asked to telephone the study coordinator and submit a stool specimen.Gastroenteritis was defined as (1) diarrhea (three or more loose stools in 24 hours); or (2) vomiting (three or more times in 24 hours); or (3) diarrhea with two or more additional symptoms; or (4) vomiting with two or more additional symptoms in 24 hours.The additional symptoms were diarrhea, vomiting, abdominal pain, cramps, fever, nausea, and blood or mucus in the stool.Cases with an obvious noninfectious cause were excluded.New cases could occur in the same person only after a 2-week symptom-free period.In total, 11,569 persons were invited and 4860 (42%) participated.The total follow-up time was 2229 person-years with 76% of participants completing the full 26 weeks of observation.During the follow-up period, 1050 case episodes occurred.This figure included multiple cases in the same persons.The overall incidence, standardized by cohort, age, and sex (according to the distribution in the Dutch population in 1999), was 0.28 episodes per person-year.The incidence of gastroenteritis was highest among children 1 -4 years of age (0.90 episodes per person-year) and lowest among children 12 -17 years of age (0.16 episodes per person-year).Adults 65 years of age or older also had a low incidence (0.19 cases per person-year).A seasonal variation in incidence was observed, with the highest rates during the winter months.

From
June 15 1999 to June 14 2000,Kuusi et al. (2003) conducted a cross-sectional survey to determine the incidence of gastroenteritis in Norway.A self-administered questionnaire was mailed to 3000 persons randomly selected from a governmental registry of all Norwegian residents.Equal proportions of participants were mailed surveys in each month of the 12-month study period to account for seasonal variation in the incidence of gastroenteritis.Participants were asked about the occurrence of gastroenteritis in the 4 weeks before the questionnaire was completed.Gastroenteritis was defined as (1) diarrhea (three or more loose stools in 24 hours), or (2) at least three of the following symptoms: vomiting, nausea, abdominal cramps, or fever $388C.Persons with chronic diarrheal illnesses were excluded.A total of 1843 persons completed the questionnaire for a 61% response rate.Gastrointestinal symptoms meeting the case definition were reported by 171 persons for an incidence of 1.2 cases per person-year.The incidence of gastroenteritis was highest among children younger than 5 years of age (12.9%) and lowest among adults 65 years of age or older (3.1%).The incidence peaked in September to October and again in December to March, with the lowest rates observed in late spring and early summer.United Kingdom and IrelandSince the early 1990s, four studies in Great Britain and one in Ireland have estimated the incidence of AGI.The first study was conducted byRoderick et al. (1995) between October 1991 and May 1992.This was a pilot study of infectious intestinal disease (IID) among persons registered at select general practices belonging to the Medical Research Council's General Practice Research Framework in England.The purpose of the study was to assess the feasibility of a larger study, subsequently conducted by Wheeler et al. (1999) and Sethi et al. (1999).Four general practices in the same area of England in both rural and urban settings were selected for a population-based cohort study.Practice registries were used to sample study participants.Random samples of individuals (in two practices) and households (in the other two practices)were drawn from the age -sex patient registers.Subjects and households were initially contacted by mail but a low response rate (31%) prompted a second round of contact during this pilot study using both mail and a follow-up telephone call, where possible.The response rate in the second round was 49% and was similar for individuals and households.A total of 192 persons were recruited for this study.Individuals and households were randomly allocated to either a 3-or 6-month follow-up group.The 6-month follow-up did not reduce compliance.Participants were asked to mail a postcard to the practice each week stating whether the individual in question or a household member had developed IID, defined as any of the following symptoms preceded by a symptom-free period of at least 3 weeks in the absence of a known noninfectious cause: (1) loose stools present for fewer than 14 days, or (2) significant vomiting for less than 48 hours that either incapacitated the patient or was forceful and accompanied by systemic symptoms.The number of loose stools required to meet the definition of IID was not specified.Similarly, significant vomiting was also not defined.If a participant experienced symptoms, the nurse was to be contacted and a stool sample was to be submitted.The incidence among individuals was higher than that among households (0.14 vs. 0.03 episodes per person-year).The overall annualized incidence of IID was 0.10 episodes per person-year.From August 1993 to January 1996, Wheeler et al. (1999) and Sethi et al. (1999) conducted a prospective medical practice-based cohort study in England to estimate the incidence infectious intestinal disease (IID) in the community.This was the follow-up study to the pilot study conducted by Roderick et al. (1995) just described.Seventy general practices were chosen from across the country to participate in the study.They were representative of general practices in England by geographical area and rural/urban location, but with fewer small and affluent practices.
studies may have reduced the estimates of the rate of AGI, thereby creating the overlap with the range of estimates provided by the prospective American studies.One anomaly should be noted in the range of estimates provided by the retrospective American studies.Sandler et al. (2000) estimated a prevalence of 3.2 episodes of loose stools or diarrhea per person-year.This estimate was more than twice as high as any other retrospective study reviewed from any country.The reasons for this outlying estimate are unknown but may reflect bias due to nonrandom participant selection resulting in a sample population not demographically representative of the US population.Furthermore, this was the only retrospective study to use the term "diarrhea" without further definition.Therefore, the case definition used in this study may have been interpreted much more liberally than those used in the other retrospective studies and may have resulted in a higher estimate.The varying definitions of AGI used in the studies we have reviewed significantly reduce their comparability.Not only are different symptoms considered in the definitions (Table However, diarrhea was most commonly defined as three or more loose or watery stools in a 24-hour period.The use of this last definition is supported by data in the scientific literature.Denno et al. (2005) showed a statistically significant trend exists between the number of stools in a 24-hour period and the presence of detectable bacterial and viral enteric pathogens.Denno et al. suggested that an incidence of fewer than three loose stools in the previous 24 hours should be used as an exclusion criterion for stool cultures.Connell et al. (1965) studied persons from industrial communities around London who were not seeking healthcare and patients attending a general medical practitioner's surgery who did not have known gastrointes- . Hall et al. (2005) used the results from the Australian national gastroenteritis survey previously described (Ashbolt et al. 2002) and excluded cases with concurrent sore throat, runny nose, sneezing, and/or cough.Using this revised case definition and weighting the results to the Australian population by age and sex, Hall estimated a national incidence of 0.92 AGI episodes per person-year.Mead et al. (1999) estimated the rate of AGI in the US population using the AGI rate calculated in the first cycle of the FoodNet population survey.Since this cycle did not collect data on the frequency of concurrent respiratory symptoms, Mead adjusted for combined gastrointestinalrespiratory illness using a value based on the Cleveland and Tecumseh studies.Mead's final estimate for the rate of AGI was 0.79 cases per person-year.As discussed, the third and fourth cycles of the FoodNet population survey collected information on respiratory symptoms among those persons with AGI.Therefore, the FoodNet estimate of AGI presented in this paper did not have to rely on the Cleveland and Tecumseh studies to adjust for concurrent respiratory illness.Data from the third and fourth FoodNet cycles indicated that 39% of those with AGI had concurrent respiratory symptoms.When these cases were excluded, the rate of AGI was estimated at 0.65 episodes per person-year.

Table 1 |
Prevalence and factors associated with reporting AGI in the month before (Scallan et al. 2005)th period.The cooperation rate was 68.2%, defined as the number of completed interviews divided by the number of completed interviews plus the number of non-interviews that involved the identification of, and contact with, an eligible respondent(Scallan et al. 2005).Data were weighted to the Australian population by age and sex and the estimated incidence of infectious gastroenteritis was 0.92 cases per person-year(Hall et al. 2005).Analysis suggested that the incidence of gastroenteritis varied by region, age, and medical history of chronic illness(Ashbolt   et al. 2002).To make the results nationally representative, data were weighted by age, sex, geographic location, household size, and the number of residential telephone lines.Using diarrhea as the outcome, defined as three or more loose stools or bowel movements in any 24-hour period, the weighted rate of diarrhea was 0.83 episodes per person-year.The incidence of diarrhea was highest among children younger than 5 years of age (8.2%) and adults 25 -44 years of age (7.8%) and lowest among adults 65 years of age and older (3.6%)(Scallan et al. 2005).definedas (1) three or more loose stools or (2) two or more episodes of vomiting in a 24-hour period.Participants who reported a chronic condition in which diarrhea or vomiting was a predominant symptom were excluded from analysis.hold,or visiting the household were excluded.Data was collected on two age groups: adults 18 years of age or older, and children 7 months to 4 years of age (through a nested survey of parents or caregivers).Within the household, the person in the age group of interest who had the most recent birthday was asked to participate.Participants were asked about the occurrence of diarrhea in the 1 month preceding the interview.Diarrhea was defined as three or more loose stools in a 24-hour period.A person with a known chronic condition in which diarrhea was a predominant symptom was excluded unless the respondent believed that the diarrhea was unrelated to the chronic condition.Estimates of the incidence of diarrhea assumed each case-patient had only one episode during the previous month.Of the 3081 annual incidence of diarrhea in Queensland for this age group was 1.49 episodes per person year, adjusted for seasonality.Since this study was conducted over a 2-month period, investigators adjusted the incidence rates for seasonality by reducing the total monthly episodes of tion.Households were randomly contacted using a directory of inhabited addresses for the study area.Households were eligible if they (1) were owner-occupied; (2) had French-speaking occupants; (3) had occupants who regularly consumed tap water; (4) had at least one child between the ages of 2 -18 years living in the household; and(5) were willing to participate.Eligible households were then randomized to a regular tap water group and a filtered water group.Households in the latter group were supplied with under-the-sink filtration units.No attempt was made to blind participants as to which treatment group they were in.The health of household participants was monitored from March 1988 to June 1989, with a 2-month gap in surveillance in July and August 1988 due to summer vacation.Family health diaries were maintained and one family member completed and returned a self-administered questionnaire for all family members every 2 weeks.Followup with a staff nurse occurred by telephone every 2 weeks.cases could occur in the same person only after 6 consecutive symptom-free days.A total of 3741 households were contacted and 606 (with 2408 individuals) were enrolled.Over the entire study period, household participation rates were 86% in the group consuming regular tap

Table 2 |
Studies providing estimates of the rate of gastrointestinal disease in developed countries a

Table 2
| (continued) Roy et al. | Rate of acute gastrointestinal illness in developed countries Journal of Water and Health | 04.Suppl 2 | 2006Table 2 | (continued) Roy et al. | Rate of acute gastrointestinal illness in developed countries Journal of Water and Health | 04.Suppl 2 | 2006

Table 2
| (continued) Roy et al. | Rate of acute gastrointestinal illness in developed countries Journal of Water and Health | 04.Suppl 2 | 2006Table 2 | (continued) Roy et al. | Rate of acute gastrointestinal illness in developed countries Journal of Water and Health | 04.Suppl 2 | 2006

Table 2
Majowicz et al. (2004)from Table3inMajowicz et al. (2004).cross-sectional community-based survey; CP, cross-sectional population-based survey; CM, cross-sectional medical practice-based survey; CBC, community-based cohort; HBC, health insurance plan-based cohort; R, recalculation of incidence rates for 1980 population; CBI, community-based intervention trial; PBC, population-based cohort; MBC, medical practice-based cohort.Roy et al. | Rate of acute gastrointestinal illness in developed countries Journal of Water and Health | 04.Suppl 2 | 2006 a b CC,